LLM-Free Q&A v0.15

Ask a question about any webpage and get the most relevant passage back — instantly, with no LLM API key. Uses BM25 relevance scoring under the hood: fast, deterministic, and completely free.

What It Does

WebPeel fetches a page, splits it into sentences, scores each sentence against your question using BM25 (the same algorithm powering search engines like Elasticsearch), and returns the top passage with a confidence score. No OpenAI key. No Anthropic key. Zero cost per query.

• ✅ No LLM API key required
• ✅ Instant — no round-trip to an LLM provider
• ✅ Completely free, unlimited use
• ✅ Deterministic — same question always returns the same answer
• ✅ Confidence score so you know how relevant the result is

CLI

# Short flag
npx webpeel "https://example.com/pricing" -q "what does the pro plan cost?"

# Long flag
npx webpeel "https://example.com/pricing" --question "what does the pro plan cost?"

# JSON output (includes confidence score)
npx webpeel "https://example.com/pricing" -q "what does the pro plan cost?" --json

# Works with any URL — news articles, docs, product pages
npx webpeel "https://en.wikipedia.org/wiki/BM25" -q "who invented BM25?"

API

# Basic quick answer
GET /v1/answer/quick?url=https://example.com/pricing&question=what+does+the+pro+plan+cost

# With curl
curl "https://api.webpeel.dev/v1/answer/quick?url=https://example.com/pricing&question=what+does+the+pro+plan+cost" \
  -H "Authorization: Bearer YOUR_API_KEY"

Query Parameters

MCP

Use the webpeel_quick_answer tool in Claude, Cursor, or any MCP client:

{
  "tool": "webpeel_quick_answer",
  "arguments": {
    "url": "https://example.com/pricing",
    "question": "what does the pro plan cost?"
  }
}

How It Works

Quick Answer uses a lightweight, purely local pipeline — no external AI calls:

1. Fetch & extract — WebPeel fetches the page and strips noise (nav, ads, footers)
2. Sentence splitting — Content is split into individual sentences using punctuation and whitespace heuristics
3. Question-type detection — Detects whether the question is a "what", "how much", "who", "when", etc. to apply appropriate boost signals
4. BM25 scoring — Each sentence is ranked against the question using BM25 term frequency–inverse document frequency
5. Boost signals — Sentences near headings, in tables, or containing numbers get score boosts for price/quantity questions
6. Top passage selection — The top-scoring sentence(s) are merged into a coherent passage and returned with a confidence score

Example

Querying a SaaS pricing page:

npx webpeel "https://example.com/pricing" -q "what does the pro plan cost?" --json

{
  "url": "https://example.com/pricing",
  "question": "what does the pro plan cost?",
  "answer": "The Pro plan costs $49/month and includes unlimited API calls, 5 team seats, and priority support.",
  "confidence": 0.87,
  "source": {
    "text": "The Pro plan costs $49/month and includes unlimited API calls, 5 team seats, and priority support.",
    "context": "Pricing — Pro",
    "score": 12.4
  }
}

SDK Usage

import { quickAnswer } from 'webpeel';

const result = await quickAnswer(
  'https://example.com/pricing',
  'what does the pro plan cost?'
);

console.log(result.answer);      // "The Pro plan costs $49/month..."
console.log(result.confidence);  // 0.87

from webpeel import WebPeel

client = WebPeel()
result = client.quick_answer(
    "https://example.com/pricing",
    "what does the pro plan cost?"
)

print(result.answer)      # "The Pro plan costs $49/month..."
print(result.confidence)  # 0.87

quick_answer vs answer

WebPeel has two answer endpoints. Choose the right one for your use case: